Early Ramping Neural Activity Reliably Precedes Self-Initiated Action But Does Not Predict It.

Experimental research on self-initiated movement has for decades relied on the same methodology: the data epochs subjected to analysis are only those culminating in a movement. This approach reliably reveals early ramping activity starting one second or more before movement, but discounts the prior probability of slow ramping. We trained powerful machine-learning classifiers in a sliding window to discriminate between EEG data epochs that either do or do not culminate in a self-initiated movement. We found early ramping signals, but they were not predictive of impending movement. Instead, the ability to correctly classify movement from no-movement epochs emerged abruptly ~100 ms before action onset, while, using the traditional approach, we replicated the spurious finding of early predictability. Our results resolve a long-standing controversy, showing the neural commitment to act is a late-stage process that aligns with results on the timing of subjective experience and the ability to inhibit self-initiated movement.